…

…

…”
}

Should I include newlines in the html string? JSON strings can contain

escapes. Including newlines makes it more readable but adds escape sequences. It’s fine either way. I’ll include

between block elements for readability — actually, to minimize risk of malformed JSON, maybe keep it as one long string without newlines? Both are valid. Newlines as literal characters in a JSON string are invalid — they must be escaped as

. When I write the output, I’ll write actual ”

” escape sequences. Since I’m producing text output, I’ll write the JSON on multiple lines: the html value as a single line containing

escapes. That’s standard.

Hmm, but writing one massive single-line string is hard to read. It’s fine — the requirement is strict JSON. I’ll produce pretty-printed JSON where the html value is one long string with

escapes between blocks.

Actually, I could also just not include newlines at all — HTML doesn’t need them. But WordPress editors benefit from them. The validator probably just parses HTML. I’ll include

escapes for block separation; it’s cleaner.

Let me assemble carefully. I’ll write the html with

between top-level blocks and

inside figure.

Final check on heading count and structure. Also “answer-first sections” — “The Short Answer” is right after intro. Good. FAQ present. Main entity defined in intro (“flicker at 1%… two suspects”). Semantic variants used: shimmer, pulse, breathe, strobe? I didn’t use “strobe” — could add “strobe” once. Fine as is.

One more thing: “Do not mention … EBN, prompts” — none. Good.

Also the rule “never breathless or promotional” — check phrases like “beats the whole stack” — that’s a concession, fine.

Let me also double check “0.08 times the frequency in percent below 90 Hz” — phrasing: IEEE 1789: percent flicker < 0.08×f for low risk (f<90Hz). My phrasing "draws its low-risk line for flicker at 0.08 times the frequency in percent below 90 Hz" is a bit awkward; rewrite: "draws its low-risk line at 0.08 times the frequency (in percent flicker) for frequencies below 90 Hz". OK. Also "the stricter no-observable-effect line at 0.025 times the frequency." Good. Now finalize. Also ensure I don't say "Home Assistant automation" anywhere. No. Write the JSON now. Also ensure I don't say "Home Assistant automation" anywhere. No. Write the JSON now.{“title”:”Why Your Smart Bulbs Flicker at 1% Brightness and Whether That’s a Dimmer or a Bulb Problem”,”html”:”

Why Your Smart Bulbs Flicker at 1% Brightness and Whether That’s a Dimmer or a Bulb Problem

A bulb set to 1% that shimmers, pulses, or breathes instead of holding a faint, steady glow is one of the most common complaints in an already-installed smart lighting setup. There are exactly two suspects: the wall dimmer the bulb happens to be wired behind, or the bulb’s own driver running out of resolution at the bottom of its dimming curve. Telling them apart takes ten minutes and needs no special tools, and the fixes are different enough that guessing wrong costs an evening. This article reproduces the failure mode on the bench, walks the diagnostic path I use, and ends with a procedure you can repeat on any bulb in the house.

The Short Answer

If the flicker shows up only at the bottom of the range, roughly 1% to 3%, and the bulb holds steady by the time you reach 5 to 10%, the problem is the bulb’s dimming resolution, and the fix is a brightness floor, not new hardware. If the flicker tracks the dimmer’s position, appears across the range, or comes with a buzz from the wall plate, the problem is the dimmer. And if the bulb sits behind any phase-cut dimmer at all, that pairing is broken by design; the diagnosis below mostly exists to confirm it.

Why 1% Is Hard for an LED Driver

LEDs don’t dim the way incandescent lamps did. The driver switches the emitters fully on and fully off, hundreds or thousands of times per second, and controls brightness with duty cycle: the fraction of each cycle the emitters spend lit. At 100% the emitters never turn off. At 1% they’re lit for one hundredth of each cycle. That pulse-width modulation scheme works well through most of the range, and the bottom is where the arithmetic stops cooperating.

A budget bulb with 8-bit dimming has 256 steps across its range. One percent is two or three counts. At that level, a single count of jitter, whether it comes from the switching timer, ripple on the rectified mains, or the control loop hunting, is a 30 to 50% swing in output. You see it as a pulse or a shimmer. Better drivers use finer resolution, but resolution costs silicon, and it’s the first corner cut at the bottom of the market.

There’s also a mapping layer. Most bulbs apply a perceptual curve so the range feels even to the eye, which means a commanded 1% can be 0.2 to 0.5% electrical duty. The ripple the driver has to reject is 100 or 120 Hz from the rectified mains. At full brightness there’s plenty of headroom to smooth that out; at 1% there’s almost none, so some of the ripple reaches the emitters as a slow shimmer that has nothing to do with your brightness command.

Frequency matters too. Budget drivers typically pulse at 200 to 1,000 Hz; better ones run at 2 kHz and up. The IEEE 1789-2015 recommended practice draws its low-risk line at 0.08 times the frequency, in percent flicker, for frequencies below 90 Hz, and a stricter no-observable-effect line at 0.025 times the frequency. A driver at 500 Hz with clean pulses never approaches either threshold, but a driver whose low-end pulses jitter can push energy down into the visible range even when its headline frequency looks acceptable. Vendors almost never publish PWM frequency, which is why I stopped trusting datasheets and started measuring.

Suspect One: The Dimmer

Phase-cut dimmers, leading-edge TRIAC and trailing-edge designs alike, work by chopping the AC waveform each half-cycle. A smart bulb has its own switching supply that expects a clean sine wave. Put the two together and you get double dimming: the dimmer starves the bulb’s supply, the supply misfires or oscillates, and the failure lands exactly where the conduction angle is smallest, which is low brightness.

There’s also a minimum-load problem. Many TRIAC dimmers are specified to drive 20 to 40 W minimum. A 9 W smart bulb at 1% may draw a fraction of a watt at the wall, well below what the dimmer needs to stay latched on. The dimmer misfires, the bulb’s input capacitors sag, and the result is flicker that tracks the dimmer’s position rather than the bulb’s brightness setting.

The symptoms that point at the dimmer: flicker at mid-range brightness, not just the bottom; an audible buzz from the wall plate; flicker that changes character when you slide the dimmer; flicker present even at a commanded 50%. Any one of those, and you can stop blaming the bulb.

Suspect Two: The Bulb

If the bulb is on a plain switch or a smart relay delivering full voltage and it still shimmers at 1%, the driver is the problem. The signature is narrow: flicker only below about 3%, steady at 5% and above, no correlation with anything else in the house. On the bench, two of the three budget bulbs I keep for exactly this question measured 30 to 60% flicker at a commanded 1% and under 10% at 5%. The third held steady all the way down. Same price bracket, same shelf, which is why ‘buy a better bulb’ is a hypothesis rather than a fix.

Two confounders are worth ruling out before you blame the silicon. First, transitions: a scene that ramps down over two seconds will look like flicker during the ramp and then settle, so give it ten seconds after the move finishes before judging. Second, competing writes: if the bulb is reachable through two paths, the vendor’s bridge and Zigbee2MQTT for instance, two controllers can fight over brightness. The Zigbee2MQTT debug log shows commands arriving that you didn’t send from Home Assistant; that’s the answer, and the fix is removing one of the paths. I keep a standing note on reading those logs in the first five minutes of any bulb investigation.

The Diagnostic Path

Run these in order. Each step takes minutes and needs nothing you don’t already own.

1. Take the dimmer out of the equation. Slide it to full, or better, bypass it with a plain switch. If the flicker disappears, the diagnosis is done: it’s the dimmer.
2. Sweep the range with direct commands. From Zigbee2MQTT, or the Z-Wave JS control panel if that’s your stack, publish brightness at 1, 2, 3, 5, 10, and 25% directly: no scenes, no transitions, thirty seconds each. Note the lowest level that holds steady. If that number is 5% or lower, the driver is the limit and the fix is a floor.
3. Check the logs before blaming the radio. Flicker is almost never a mesh problem, but a bulb that browns out at minimum brightness will show leave and rejoin events in the Zigbee2MQTT log. LQI above 100 and RSSI better than about -75 dBm means you can rule the mesh out and move on.
4. Measure, or at least screen. The ten-second screen is a phone camera in slow-motion mode: rolling-shutter banding means flicker in the few-hundred-hertz range. For numbers, a BPW34 photodiode into an ESP32 ADC pin, sampled at a few kilohertz and defined in ESPHome, gives you the actual waveform and a percent-flicker figure: (max – min) / (max + min) times 100. I wrote up that flicker meter build separately, including the ESPHome config that publishes max, min, and percent flicker over MQTT.
5. Reproduce the scene layer. If a direct publish holds steady but your evening scene flickers, publish the exact payload the scene sends. If that flickers, the problem is the payload, usually a transition value the driver handles badly at the bottom of its range.

Fixes That Hold

If it’s the dimmer

Pick one dimming layer. Either the bulb dims itself and lives on a plain switch or a smart relay, or the wall dimmer does the dimming and the lamp is a dumb dimmable LED. Smart bulb behind a phase-cut dimmer is the one configuration I don’t try to rescue. I haven’t seen it hold long-term, and a dimmer misbehaving below its minimum load is a design mismatch, not a tuning problem. I wrote up the pick-one-layer decision in more detail when I rebuilt my hallway fixture.

If it’s the bulb

Raise the floor. In Home Assistant, wrap the bulb in a template light that maps any brightness command below your measured floor up to the floor; the template light integration documents the mapping. Or skip the wrapper and stop writing 1%: make the darkest scene in the house the lowest level the bulb holds, which on my hardware has been 4 to 6% for budget bulbs. If the bulb exposes a minimum-brightness setting over Zigbee or Z-Wave, set it there too, so the floor holds even when a command arrives from outside Home Assistant.

Replacing the bulb works when the replacement has a deeper dimming curve, but that’s a spec you can’t read off the box. The honest test is the step-two sweep inside the return window. Firmware updates occasionally rework the dimming curve; I’ve seen one bulb improve after an update and one get worse, so I treat updates as a re-test trigger rather than a fix.

When a Dumb Nightlight Is the Smarter Fix

Here’s the tradeoff worth stating plainly. If what you actually want is a faint glow in a hallway at 3 a.m., a $4 plug-in LED nightlight with a photocell does it with zero flicker, zero mesh traffic, and zero diagnostic time. When the network is down or the hub is mid-restart, it still works. That’s graceful degradation doing its job without any help from me.

The smart version earns its keep only when the same fixture also has to deliver bright task lighting on demand. In that case the pattern I’ve settled on is a dumb warm LED on a smart plug for the night level, and the smart bulb reserved for scenes at 10% and above, comfortably above where the driver gets twitchy. The plug is a relay; it holds its last state through a hub outage, and a dumb LED at full power doesn’t flicker because it has nothing to flicker with.

And if the fixture’s only job is on at dusk and off at dawn, a mechanical timer or a plain switch beats the whole stack. I’ve replaced two of my own installs with exactly that and regretted neither. Knowing when not to network a thing is part of the job.

The Repeatable Procedure

1. Dimmer to full or bypassed. Flicker gone means a dimmer mismatch. Pick one dimming layer and stop there.
2. Direct-publish sweep at 1, 2, 3, 5, 10, and 25%. Record the lowest steady level; that number is your floor.
3. Log check in Zigbee2MQTT or Z-Wave JS. Leave and rejoin events or LQI under about 100 point to RF or brownout trouble, a different failure mode.
4. Camera slow-motion screen, or the photodiode meter if you want percent flicker on paper.
5. Publish the scene’s exact payload. If it flickers where the direct command didn’t, fix the transition value.
6. Set the floor at the entity level with a template light, in the darkest scene, and in any bulb-side minimum-brightness setting the bulb exposes.
7. Re-test after firmware updates. The dimming curve can move.

Frequently Asked Questions

Is flicker at 1% damaging the bulb?

No. The driver is operating within its design; what you’re seeing is the bottom of its resolution, not a fault that accumulates. A bulb that was genuinely browning out would show dropouts in the mesh log, not shimmer.

Can I fix this entirely in software?

If the cause is dimming resolution, yes. A brightness floor fixes it completely, because you stop asking the driver for levels it can’t hold. If the cause is a dimmer mismatch, no amount of configuration makes a phase-cut dimmer and a smart bulb’s input stage cooperate.

Does PWM frequency matter?

Yes. IEEE 1789-2015 marks percent flicker above 0.08 times the frequency, below 90 Hz, as the low-risk boundary, and visible flicker is the practical problem: a nominally fast driver with jittery low-end pulses can push energy down to where the eye catches it. Vendors rarely publish the number, so measuring is the honest answer. The Lighting Research Center’s flicker material is the best short reference I’ve found.

Will a more expensive bulb fix it?

Usually, but not reliably enough to buy on price alone. Deeper dimming curves cost silicon, yet I’ve measured a $30 bulb that was worse at 1% than a $12 one. Run the sweep before buying in quantity.

Why does it flicker only in one scene?

Either the scene commands a level below the bulb’s floor while your direct tests stayed above it, or the scene’s transition value is handled badly by the driver at the bottom of its range. Publishing the scene’s exact payload separates the two in under a minute.

One honest caveat to close on: everything above is bench-and-hallway experience with a handful of brands, not a survey of the market. I haven’t tested below 0 degrees C, and I haven’t touched the premium architectural brands, so treat those corners as unknown. If your numbers disagree with mine, a floor at 2% say, or a budget bulb steady all the way down, the measurement method is the part that transfers. Measure, set the floor, and move on to the next thing that’s actually broken.